Background and purpose: Many prostate cancer patients undergoing pelvic lymph node dissection (PLND) have no sign of lymph node invasion (LNI) during final pathological assessment. To improve preoperative staging accuracy, we developed the Oslo model, which estimates the risk of LNI based on clinical, histopathological, and magnetic resonance imaging (MRI) variables.
Patients/materials and methods: We utilized data from 903 prostate cancer patients treated at Oslo University Hospital (OUS) to develop the model using Bayesian logistic regression. The Oslo model was validated with data from 189 patients at IRCCS Ospedale San Raffaele (HRS), 157 from St. Olav's Hospital, and 231 from OUS. We assessed its performance against the Memorial Sloan Kettering Cancer Centre (MSKCC) and Briganti 2019 nomograms using metrics like AUC, R², decision curve analysis, and calibration plots.
Results: The Oslo model outperformed Briganti 2019, demonstrating a higher net benefit and a 10% reduction in interventions at a 7% cutoff. Key variables included clinical T stage on MRI, Prostate Specific Antigen (PSA), prostate volume, International Society of Urological Pathology grade group, and maximum lesion length on MRI. Validation showed strong reliability in the OUS and HRS cohorts but weaker performance in the St. Olav's cohort. The AUCs were 77% for the Oslo model, 74% for Briganti 2019, and 66% for MSKCC. Limitations include small and heterogeneous validation cohorts.
Interpretation: The Oslo model enhances predictive performance in intermediate- and high-risk patients using easily accessible clinical and MRI data, potentially reducing unnecessary PLND interventions and assisting clinicians in treatment decision-making.